Descriptions:
This Veteran AI tutorial (presented in Mandarin) demonstrates a “Grid Method” for generating multiple storyboard shots of the same scene with guaranteed character and environment consistency, combining Qwen Image Edit 2512 with LTX Two for final video output. Rather than using reference-based generation—which can introduce subtle inconsistencies, especially with open-source models—the approach generates all storyboard panels simultaneously within a single image, ensuring coherence because all shots share one sampling pass.
The workflow has three stages. First, Qwen Image Edit version 2512 (specifically not the 2511/Plus version, which is shown side-by-side to produce noticeably weaker multi-panel layouts) generates a grid of four shots at 1440×720. Second, the “Unblur Anything” LoRA—compatible with Qwen 2509 and 2511 models—upscales and sharpens each panel to 1920×1080, recovering facial detail that gets lost when four images share limited resolution. Third, each upscaled frame is fed into the low-VRAM LTX Two GGUF workflow to generate individual 1280×720 video clips.
A practical compatibility issue is flagged: Kijai’s newly uploaded VAE version causes gray or corrupted output in this pipeline, so the older VAE version must be selected. The method is presented as more reliable than reference-based generation for multi-shot sequences where character and set continuity are critical. All workflows are available on RunningHub.
📺 Source: Veteran AI · Published January 16, 2026
🏷️ Format: Workflow Case Study
